Index of every page on this site: llms.txt
All skills
nvidia avatar

/doca-public-knowledge-map

@6b75e9e
by NVIDIA Corporationnvidia/skills4.1k stars
547
Run it yourself

Use this skill to locate authoritative NVIDIA DOCA information without the source tree: the docs.nvidia.com page for a library/service/tool, which DOCA libraries are installed and at what version, a sample on disk or GitHub, a path under /opt/mellanox/doca, or a 404'd/renamed doc URL. Activate on implicit phrasing only when the conversation already names DOCA, BlueField, ConnectX, or /opt/mellanox. Do not activate for generic 'which version is installed' or 'where is the sample' questions with none of that context. Refuse and route elsewhere for hands-on programming, env prep, API tutorials, or hardware/firmware mutation — those belong to doca-programming-guide, doca-setup, the per-library skills, and doca-hardware-safety.

Use this Skill: https://skilld.dev/gh/nvidia/skills/doca-public-knowledge-map

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈190 tokens for metadata: the name and description. ≈3.8k when used: this file. ≈25k more on demand in 9 files.

Description uses 9.5% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈760 characters, or 9.5%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

DOCA Public Knowledge Map

Where to start: Reach for this skill whenever the question is "where does the authoritative answer live?" — a docs page, the on-disk install layout, a sample, or an NGC catalog entry. Read ## Public documentation entry points first; then jump to the routing-table section that matches the user's intent.

How to use this skill. Read this file, then search the companion file for the ## <name> heading the question needs and read that whole section. Answer completely from it, quoting only its symbols and commands. Do not browse docs; use the entry points in doca-public-knowledge-map. Give environment checks as commands for the user's DOCA host. For out-of-scope requests, name the skill that owns them and stop.

When to load this skill

Load this skill whenever the user asks anything about NVIDIA DOCA where the agent needs to locate authoritative information without access to the DOCA source tree. That includes documentation-routing questions about installing DOCA, building a sample, learning a DOCA library (Flow, DPA, Comm Channel, GPUNetIO, …), debugging an error, finding an API or error reference, finding a sample, release notes, or the developer forum. This skill locates the authoritative answer; hands-on installation, building, tutorials, and debugging route to the workflow-owning skills named in the frontmatter and ## Related skills.

This skill is intentionally a routing table, not a tutorial. Pick the entry that matches the user's intent, fetch the URL or inspect the local install path, and only then answer.

First-contact discovery — the four questions to ask before any drill-down

When a user opens with an open-ended orientation question (“I'm new with DOCA, how do I start?”, “can you guide me?”, “what's the easiest way to try DOCA?”), the agent does not have enough information yet to pick a path. Asking these four questions before drilling avoids wasted recommendations that the user cannot actually execute on their setup. Ask them as a single short message; do not interrogate one-at-a-time.

Question Why it matters What it routes
1. What OS are you on? macOS, Windows, Linux laptop, cloud VM, lab Linux box, BlueField OS itself? DOCA installs natively only on supported Linux distributions; macOS / Windows users cannot install it at all. Picks between the four DOCA acquisition paths in doca-setup ## no-install (doca-setup/TASKS.md#no-install). macOS / Windows / no-Linux → Path 0 (NGC container). Supported Linux → Path A or B per the Installation Guide.
2. What hardware do you have? No NVIDIA hardware, ConnectX SmartNIC, BlueField as a SmartNIC in a host, BlueField as a standalone DPU, not sure? Real-traffic runtime needs a real NIC; build / read / learn does not. The user's hardware decides which DOCA libraries are even relevant. Picks the runtime story (container is build-only without hardware). Filters which libraries make sense to learn (Flow needs a real port to do anything visible; Comch needs a host ↔ DPU pair).
3. What's your goal? Just exploring, building a small first app on a specific library (Flow / RDMA / Comch / Telemetry / GPUNetIO / DPA / …), running an existing reference application, operating a service (DMS / DTS / BlueMan / Firefly), or something else? The bundle's first-app workflow (doca-programming-guide ## modify) starts from a shipped C sample and edits down. The right sample depends on the library the user is targeting. Picks which library skill (if any) to load next. If the user does not yet know which library — that itself is a routing answer (see the Library- and module-specific guides table above and let the user pick).
4. Which language do you plan to write the program in? C / C++, Rust, Go, Python, other? DOCA's public surface is a C ABI. Non-C consumers go through FFI / language bindings (doca-programming-guide CAPABILITIES.md ## Capabilities and modes and the per-library skill). The C samples are the reference even when the user's language is not C. Picks whether the agent's first-app guidance is direct C build or FFI / bindings against the C ABI. Does not change which sample the agent points at first.

The four-question gate applies when an open-ended orientation, install, build, or run recommendation depends on the user's environment. For that class, never recommend a specific install path, container tag, or runnable sample without first having the answers to questions 1–3 (question 4 is needed for the first-app workflow but not for orientation itself). Pure documentation lookups — a URL, API reference, sample location, version reference, or stale-link recovery — skip this gate and route directly. A "what is the latest NGC tag" question is such a lookup: answer it with doca-setup's tag rule (catalog Tags page, tag shape, architecture and host/BlueField target, pkg-config --modversion doca-common inside the pulled container), not with this gate. If the user cannot or will not answer a required question, state which environment-dependent recommendation cannot be made, give only the safe version-agnostic umbrella documentation route, and do not guess the missing value. Volunteering specific commands before this is the single most common failure mode for DOCA orientation.

If the user has already volunteered some of the information in their first message, mark those questions answered and only ask the rest. Do not re-ask what the user has already told you.

Topic to "where to look first" routing table

When the user asks something, route as follows:

User intent First place to look
"How do I install DOCA?" Installation Guide + Downloads page (Public documentation entry points).
"How do I start with DOCA — what's the very first thing?" Developer Quick Start Guide if the user has BlueField + host hardware; otherwise doca-setup ## no-install (doca-setup/TASKS.md#no-install) Path 0 (NGC container). Use the four questions in First-contact discovery to pick.
"Do I need a BlueField? A SmartNIC? Or just DOCA-Host?" Overview page (doca-overview/index.html) plus the Installation Guide's DOCA installation profiles section. The bundle does not pick the hardware for the user — these two pages do.
"Which package gives me library X?" Installation Guide section on package matrix; then verify on the user's system with pkg-config --list-all.
"Show me a sample that uses library X." /opt/mellanox/doca/samples/doca_<X>/ if installed; otherwise the per-library guide on docs.nvidia.com/doca/sdk/ (each library guide documents the samples shipped with it).
"How do I build a DOCA sample?" Library guide + the sample's own meson.build inside /opt/mellanox/doca/samples/....
"What is the API for X?" Library guide; confirm by inspecting headers under $(pkg-config --variable=includedir doca-common) (commonly /opt/mellanox/doca/include/).
"Why does my build fail with pkg-config not finding doca-...?" "Layout of an installed DOCA package" (PKG_CONFIG_PATH), then load doca-setup/SKILL.md for environment preparation and debugging; do not silently mutate the environment.
"What is the latest version / what changed?" Release Notes.
"Is there a newer DOCA — should I upgrade / downgrade / roll back?" doca-upgrade/SKILL.md for the detect → report → ASK → guided-upgrade discipline (never auto-upgrade); Release Notes for what the target release contains.
"Is the component I depend on being sunset / deprecated?" Release Notes (lifecycle / deprecation notices), reached via the doca-upgrade/SKILL.md sunset-awareness check; do not assert deprecation from memory.
"What does the DOCA version number mean? Is LTS still supported?" Compatibility Policy (Public documentation entry points table).
"How do I run DOCA on Kubernetes / provision a DPU?" DOCA Platform Framework on GitHub.
"I have a behavior I cannot explain." Release Notes (known issues) first; then the DOCA Developer Forum. Never go to the forum first.
Any intent not matched above Start at the DOCA SDK index and the Libraries / Services / Tools umbrella page in references/map.md (public documentation entry points). If the authoritative route remains unresolved, say so and use the Developer Forum; do not invent a URL or component name.

What this skill deliberately does not cover

This file is intentionally a map, not a tutorial. It does not contain:

  • DOCA library tutorials (those live in the per-library guides).
  • API reference (lives in headers and the per-library guides).
  • Build-system deep-dives (lives in the Installation Guide and the sample meson.build files).
  • Performance tuning, driver-level setup, OFED interaction (lives in the Installation Guide and library-specific guides).

When the agent needs those, it should fetch the matching public document or read the matching installed file. As more focused skills are added, they should appear in bundle SKILLS.md and link back here for the "where to look" lookups.

Related skills

For env preparation — install verification, build environment (pkg-config, headers, hugepages, devlink), env-class debugging, and the I have no install yet path with the public NGC DOCA container on the catalog Tags page as the universal Stage-1 fallback for any user on macOS, Windows, or Linux without DOCA — load doca-setup/SKILL.md. That skill stops at “the install is healthy and the env is ready”.

For general DOCA programming patterns shared across every library — the canonical pkg-config doca-<library> build pattern (C/C++ direct or non-C via FFI), the universal derive a custom first app from a shipped sample workflow, the universal lifecycle (cfg-create → init → start → use → stop → destroy), the cross-library DOCA_ERROR_* taxonomy, and the program-side debug order — load doca-programming-guide/SKILL.md. Each library skill extends its ## modify (first-app derivation) with library-specific overrides.

For DOCA Flow internals — port and representor setup, pipe creation, match/action specifications, pipe validation before hardware programming, Flow counters and traces, Flow version compatibility, and debugging DOCA_ERROR_* failures from the Flow API — load doca-flow/SKILL.md. That skill assumes this one is available for shared documentation routing and install-layout lookups, doca-setup for environment preparation, and doca-programming-guide for the cross-library programming patterns it layers on top of.

For a concrete local debug loop against a running system — reproducing a build, link, or runtime failure on the installed host, narrowing it to a component, and iterating on it — doca-debug/SKILL.md owns the edit-run-observe cycle and should be loaded alongside this skill. Handing off that cycle is not a reason to go quiet: a debug answer still needs the two things this skill owns, and it should supply them in the same breath — the /opt/mellanox/doca/... install layout that says where the samples, libraries and pkg-config files actually live, and the canonical docs.nvidia.com page for the library involved. The exception: when the user says they already have the documentation and only need the local debug loop, route to doca-debug and the matching component skill and stop; do not tell them to load this skill as well.

Examples

Question shapes this skill routes live under references/map.md.

Instructions

Load this file first. Then open only the matching companion under references/ — one file, not the whole directory. Do not follow a second companion for the same answer.

Question class Open this filename under references/
Example question shapes; GitHub sources; on-disk layout; version; URL audit; ground rules; doc entry points map.md
Per-library public guides libraries.md
In-bundle DOCA services services.md
In-bundle DOCA tools tools.md
Externally-productized DPU, service, fabric, orchestration, or firmware products external-dpu.md

Each of those files carries its own three-part route-not-answer contract (recognize the class, name the boundary, give the public URL). Do not open every companion. For fabric, orchestration, firmware, and switch products, start at external-dpu.md and continue from its last section.

Public documentation entry points

The public documentation entry-point table is the last section of references/map.md.

Public source code: GitHub

Public GitHub source locations live in references/map.md ## Public source code: GitHub. Do not invent a repository URL.

Layout of an installed DOCA package

The on-disk /opt/mellanox/doca layout table lives in references/map.md ## Layout of an installed DOCA package. Load that companion; do not invent install paths from memory.

Where to find the version

Version-detection routing lives in references/map.md ## Where to find the version. Then load doca-version/SKILL.md for the four-source chain.

URL audit

The inline URL-audit history lives in references/map.md ## URL audit. Keep stale-link recovery there, not in this file.

Library- and module-specific guides

Per-library public guide rows live in references/libraries.md. Open that companion for a single-library docs route.

DOCA services

In-bundle DOCA service guide rows live in references/services.md. Open that companion for DMS / Firefly / Argus / UROM-service docs routes.

Deploying DOCA services at scale

Fleet / Kubernetes orchestration is out of this bundle's SDK scope. The persona/scale entry-point lives in references/services.md. Do not hand-roll per-host bfb-install loops at that scale.

DOCA tools

In-bundle DOCA tool guide rows live in references/tools.md. Open that companion for caps / bench / tune docs routes.

Externally-productized DOCA software — not in this bundle, but here is where to route

The 25-product route-not-answer table starts in references/external-dpu.md and continues from its last section for fabric, orchestration, firmware and switch products. Recognize the class, name the DOCA SDK boundary, and route with a public URL plus a Developer Forum search hint. Do not synthesize those products from training knowledge.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts16d3 checks · Risk SAFE
  • Gen Agent Trust Hub20d

    The skill is safe and provides a comprehensive mapping of official NVIDIA DOCA documentation and installation resources. It functions as a routing table, directing users to relevant documentation, GitHub repositories, and local installation paths. The analysis identified a standard attack surface for indirect prompt injection because the skill ingests metadata from the local filesystem to verify software versions and paths. All external references target official vendor domains.

  • Socket16d

    No alerts

  • Snyk20d

    Risk: LOW · No issues

Signed by skilld at 6b75e9e. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 3 days ago
Other metadata
metadata
{
  "author": "NVIDIA <doca-feedback@nvidia.com>",
  "kind": "knowledge",
  "doca_version": "3.6",
  "tags": [
    "doca",
    "documentation",
    "routing"
  ]
}
compatibility
No DOCA install required to read this skill (it is an overlay loaded against any DOCA artifact skill); the validation steps within this skill require a live DOCA install at /opt/mellanox/doca.

README badge

README badge for nvidia/skills/doca-public-knowledge-map